1 / 16

Score-P hands-on: NPB-MZ-MPI / BT (cont . )

Score-P hands-on: NPB-MZ-MPI / BT (cont . ). Performance Analysis Steps. 0.0 Reference preparation for validation 1.0 Program instrumentation 1.1 Summary measurement collection 1.2 Summary analysis report examination 2.0 Summary experiment scoring

lynnea
Download Presentation

Score-P hands-on: NPB-MZ-MPI / BT (cont . )

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Score-P hands-on:NPB-MZ-MPI / BT (cont.)

  2. Performance Analysis Steps 0.0 Reference preparation for validation 1.0 Program instrumentation 1.1 Summary measurement collection 1.2 Summary analysis report examination 2.0 Summary experiment scoring 2.1 Summary measurement collection with filtering 2.2 Filtered summary analysis report examination 3.0 Event trace collection 3.1 Event trace examination & analysis

  3. Warnings and Tips Regarding Tracing • Traces can become extremely large and unwieldy • Size is proportional to number of processes/threads (width), duration (length) and detail (depth) of measurement • Traces containing intermediate flushes are of little value Uncoordinated flushes result in cascades of distortion • Reduce size of trace • Increase available buffer space • Traces should be written to a parallel file system • /work or /scratch are typically provided for this purpose • Moving large traces between file systems is often impractical • However, systems with more memory can analyze larger traces • Alternatively, run trace analyzers with undersubscribed nodes

  4. Measurement Configuration: scorep-info • Score-P measurements are configured via environmental variables: % scorep-info config-vars --full SCOREP_ENABLE_PROFILING Description: Enable profiling [...] SCOREP_ENABLE_TRACING Description: Enable tracing [...] SCOREP_TOTAL_MEMORY Description: Total memory in bytes for the measurement system [...] SCOREP_EXPERIMENT_DIRECTORY Description: Name of the experiment directory [...] SCOREP_FILTERING_FILE Description: A file name which contain the filter rules [...] SCOREP_METRIC_PAPI Description: PAPI metric names to measure [...] SCOREP_METRIC_RUSAGE Description: Resource usage metric names to measure [... More configuration variables ...]

  5. BT-MZ Trace Measurement Collection... • Re-run the application using the tracing mode of Score-P • Edit scorep.lltoadjustconfiguration • Submit job • Separate trace file per thread written straight into new experiment directory ./scorep_bt-mz_C_32x8_trace export SCOREP_EXPERIMENT_DIRECTORY=scorep_bt-mz_C_32x8_trace exportSCOREP_FILTERING_FILE=../config/scorep.filt exportSCOREP_ENABLE_TRACING=true exportSCOREP_ENABLE_PROFILING=false export SCOREP_METRIC_PAPI=PAPI_FP_OPS export SCOREP_TOTAL_MEMORY=50M % llsubmitscorep.ll

  6. Advanced Measurement Configuration: Metrics • Recording hardware counters via PAPI • Also possible to record them only per rank • Recording operating system resource usage export SCOREP_METRIC_PAPI=PAPI_L2_TCM,PAPI_FP_OPS export SCOREP_METRIC_PAPI_PER_PROCESS=PAPI_L3_TCM export SCOREP_METRIC_RUSAGE_PER_PROCESS=ru_maxrss,ru_stime Note:Additional memory is needed to store metric values. Therefore, you may have to adjust SCOREP_TOTAL_MEMORY,for example as reported using “scorep-score –c”

  7. Advanced Measurement Configuration: Metrics • Available PAPI metrics • Preset events: common set of events deemed relevant and useful for application performance tuning • Abstraction from specific hardware performance counters,mapping onto available events done by PAPI internally • Native events: set of all events that are available on the CPU(platform dependent) Run this on the compute nodes! % papi_avail % papi_native_avail • Note: • Due to hardware restrictions • number of concurrently measured events is limited • there may be unsupported combinations of concurrent events

  8. Advanced Measurement Configuration: Metrics • Note: • Not all fields are maintained on each platform. • Check scope of metrics (per process vs. per thread) • Available resource usage metrics • % man getrusage • [... Output ...] • structrusage { • structtimevalru_utime; /* user CPU time used */ • structtimevalru_stime; /* system CPU time used */ • long ru_maxrss; /* maximum resident set size */ • long ru_ixrss; /* integral shared memory size */ • long ru_idrss; /* integral unshared data size */ • long ru_isrss; /* integral unshared stack size */ • long ru_minflt; /* page reclaims (soft page faults) */ • long ru_majflt; /* page faults (hard page faults) */ • long ru_nswap; /* swaps */ • long ru_inblock; /* block input operations */ • long ru_oublock; /* block output operations */ • long ru_msgsnd; /* IPC messages sent */ • long ru_msgrcv; /* IPC messages received */ • long ru_nsignals; /* signals received */ • long ru_nvcsw; /* voluntary context switches */ • long ru_nivcsw; /* involuntary context switches */ • }; • [... More output ...]

  9. Advanced Measurement Configuration: MPI • Record only for subset of the MPI functions events • All possible sub-groups • cg Communicator and group management • coll Collective functions • env Environmental management • err MPI Error handling • ext External interface functions • io MPI file I/O • misc Miscellaneous • perfPControl • p2p Peer-to-peer communication • rma One sided communication • spawn Process management • topo Topology • type MPI datatype functions • xnonblock Extended non-blocking events • xreqtest Test events for uncompleted requests export SCOREP_MPI_ENABLE_GROUPS=cg,coll,p2p,xnonblock

  10. Advanced Measurement Configuration: CUDA • Record CUDA events with the CUPTI interface • All possible recording types • runtime CUDA runtime API • driver CUDA driver API • gpu GPU activities • kernel CUDA kernels • idle GPU compute idle time • memcpy CUDA memory copies (not available yet) % export SCOREP_CUDA_ENABLE=gpu,kernel,idle % mpiexec -n 4 ./bt-mz_B.4 NAS Parallel Benchmarks (NPB3.3-MZ-MPI) - BT-MZ MPI+OpenMP Benchmark [... More application output ...]

  11. Score-P User Instrumentation API • Can be used to mark initialization, solver & other phases • Annotation macros ignored by default • Enabled with [--user] flag • Appear as additional regions in analyses • Distinguishes performance of important phase from rest • Can be of various type • E.g., function, loop, phase • See user manual for details • Available for Fortran / C / C++

  12. Score-P User Instrumentation API (Fortran) • Requires processing by the C preprocessor #include "scorep/SCOREP_User.inc" subroutine foo(…) ! Declarations SCOREP_USER_REGION_DEFINE( solve ) ! Some code… SCOREP_USER_REGION_BEGIN( solve, “<solver>", \ SCOREP_USER_REGION_TYPE_LOOP ) do i=1,100 [...] end do SCOREP_USER_REGION_END( solve ) ! Some more code… end subroutine

  13. Score-P User Instrumentation API (C/C++) #include "scorep/SCOREP_User.h" void foo() { /* Declarations */ SCOREP_USER_REGION_DEFINE( solve ) /* Some code… */ SCOREP_USER_REGION_BEGIN( solve, “<solver>", \ SCOREP_USER_REGION_TYPE_LOOP ) for (i = 0; i < 100; i++) { [...] } SCOREP_USER_REGION_END( solve ) /* Some more code… */ }

  14. Score-P User Instrumentation API (C++) #include "scorep/SCOREP_User.h" void foo() { // Declarations // Some code… { SCOREP_USER_REGION( “<solver>", SCOREP_USER_REGION_TYPE_LOOP ) for (i = 0; i < 100; i++) { [...] } } // Some more code… }

  15. Score-P Measurement Control API • Can be used to temporarily disable measurement for certain intervals • Annotation macros ignored by default • Enabled with [--user] flag #include “scorep/SCOREP_User.inc” subroutine foo(…) ! Some code… SCOREP_RECORDING_OFF() ! Loop will not be measured do i=1,100 [...] end do SCOREP_RECORDING_ON() ! Some more code… end subroutine #include “scorep/SCOREP_User.h” void foo(…) { /* Some code… */ SCOREP_RECORDING_OFF() /* Loop will not be measured */ for (i = 0; i < 100; i++) { [...] } SCOREP_RECORDING_ON() /* Some more code… */ } Fortran (requires C preprocessor) C / C++

  16. Further Information Score-P • Community instrumentation & measurement infrastructure • Instrumentation (various methods) • Basic and advanced profile generation • Event trace recording • Online access to profiling data • Available under New BSD open-source license • Documentation & Sources: • http://www.score-p.org • User guide also part of installation: • <prefix>/share/doc/scorep/{pdf,html}/ • Contact: info@score-p.org • Bugs: support@score-p.org

More Related